An AI-native platform team / Our direction
AI does the legwork. Engineers make the call.
Our direction is for operator agents to do most routine work, with humans approving and every action leaving an audit trail. Models propose, policy decides, a deterministic executor acts, and verification closes the loop. Production changes need human approval.
Capabilities and their status
Only AI-assisted readiness reviews are in use, within the review service. None of these AI capabilities is generally available as a product. The rest describes work in development or planned work.
AI-assisted readiness reviews
In use
For the review service, we use AI tools to collect, summarise and cross-check evidence. An engineer verifies every finding before it reaches you. Customer data goes to AI tools only with your written consent.
Upgrade prep agent
In development
Our AI DevOps agent for EKS is in development to find deprecated APIs, check add-on, node and Karpenter compatibility, and draft the upgrade plan and a pre/post health checklist for human approval.
Explain this failure
In development
An AI SRE for Kubernetes capability in development: read the deployment timeline, build/deploy events and logs, then explain why a deploy failed with a suggested fix.
Monthly operations receipt
Planned
A planned record of every change: who or what requested it, who approved it, what ran and the verified result.
Restore drills and audit evidence
Planned
Planned scheduled restore tests with measured recovery time and data loss, plus infrastructure evidence from real cluster state.
An interface for your coding agents
Planned
A planned MCP server for Kubernetes deployments and Claude Code / Codex plugin over the same operations: read-only first (deploy status, logs, explain), then staging writes, then production changes that require approval.
Guardrails we won't compromise
These are the boundaries for our review service and requirements for the agent roadmap.
- Human approval for production changes.
- Scoped, revocable access.
- A full audit trail of requests, approvals, execution and verified results.
- Replaceable models, with no lock-in to one AI vendor.
- Your data is not used to train models.
- AI suggestions are labelled as such and verified by an engineer.
Start with a readiness review
Start with a read-only review and engineer-verified findings, or discuss the operating needs you would bring as a design partner.
Questions before you start
Is this an AI SRE tool?
Our AI SRE for Kubernetes direction focuses on explaining failures and preparing operational work for engineer review. Today, AI assistance is in use within our readiness review service; the failure explanation capability is in development. We do not offer a generally available autonomous SRE product, and production changes require human approval.
Does AI change production on its own?
No. Production changes require human approval. Our agent roadmap follows a controlled loop: models propose, policy decides, a deterministic executor acts and verification closes the loop.
Which models do you use?
We choose the best available models for the task and keep them replaceable. Currently, Anthropic Claude and OpenAI Codex models are used internally; we do not lock the operations workflow to one AI vendor.
Is my data used for training?
No. We only use providers and settings that do not train on your data. Customer data goes to AI tools only with your written consent covering the tool and intended data. An engineer verifies every finding before delivery.
What is live today?
AI-assisted readiness reviews are in use as part of the review service. The upgrade prep agent and Explain this failure are in development. Operations receipts, restore drills, audit evidence and the coding-agent interface are planned. None of these AI capabilities is generally available as a product.
Can I join early?
We welcome conversations with design partners about real operating needs and early access. Contact us to discuss fit, scope and availability; planned capabilities are not available today.